Datenkrake open data

Read-only access to the data the Datenkrake collects about German smart-home YouTube channels: videos, view/like/comment counts over time, titles and thumbnails including A/B tests, transcripts, sponsor positions and the analyses built on them.

SQL

Any Postgres client works: psql, DBeaver, DataGrip, pgAdmin, Metabase, Grafana, Excel/Power BI, pandas, R, DuckDB.

Host	`open.datenkrake.gastspiel.rath.li`
Port	`5432`
Database	`datenkrake`
User / password	`open` / `open`
Schema	`open` (the default search path)

psql "postgresql://open:open@open.datenkrake.gastspiel.rath.li:5432/datenkrake?sslmode=require"

-- daily views of the newest videos
SELECT v.title, s.observed_at::date AS day, max(s.views) AS views
FROM videos v JOIN video_stats s ON s.video_id = v.id
WHERE v.published_at > now() - interval '7 days'
GROUP BY 1, 2 ORDER BY 1, 2;

# DuckDB
INSTALL postgres; LOAD postgres;
ATTACH 'host=open.datenkrake.gastspiel.rath.li user=open password=open dbname=datenkrake' AS dk (TYPE postgres, READ_ONLY);
SELECT count(*) FROM dk.open.video_stats;

Every table has a comment explaining it: \d+ open.* in psql, or the comment column in your client.

HTTP API

The same tables as JSON or CSV, through PostgREST. The OpenAPI description is at /api/.

curl 'https://open.datenkrake.gastspiel.rath.li/api/channels?select=id,title'
curl 'https://open.datenkrake.gastspiel.rath.li/api/videos?channel_id=eq.UC…&order=published_at.desc&limit=20'
curl 'https://open.datenkrake.gastspiel.rath.li/api/video_stats?video_id=eq.…&order=observed_at' -H 'Accept: text/csv'

Responses are capped at 10 000 rows; page with limit and offset, or use SQL for bulk work.

Tables

`channels`, `channel_stats`	tracked channels, subscriber and view counts over time
`videos`, `video_stats`	videos, view/like/comment counts over time (one row per poll)
`video_titles`, `thumbnails`, `video_variants`	every title and thumbnail seen, and which ones YouTube served together
`video_descriptions`, `transcripts`	description history, timed transcripts
`video_heatmap`, `video_ad_segments`, `video_storyboard`, `video_music`	most-replayed curve, sponsor positions, preview geometry, credited songs
`channel_posts`, `comment_counts`	community posts, number of comments collected per video
`analysis_*`, `thumbnail_features`	transcript, comment, LLM and thumbnail analyses, precomputed graphs

Not included: the comments themselves and anything that names individual commenters.

Limits

Queries stop after 60 seconds; idle sessions close after 15 minutes.
40 SQL connections in total, shared by everyone. Disconnect when you are done.
The HTTP API allows 10 requests per second per IP.
The data comes from a replica and is usually a few seconds behind. Long queries can be cancelled when the replica applies changes; just run them again.
Since 24 August 2026 YouTube counts a view from the first frame. View counts of videos uploaded after that are not comparable with earlier ones.